Vehicle track recovery method, computer equipment and storage medium

Abstract

The invention relates to a vehicle track recovery method, computer equipment and a storage medium. Single-source track data sets for at least two types of traffic perception data in GPS data, vehicle snapshot data and toll station gate data are constructed respectively; performing fusion processing on each single-source trajectory data set to obtain a first multi-source trajectory data set; cross-data-source matching is carried out on vehicle identities corresponding to trajectory fragments in the first multi-source trajectory data set based on road network space-time constraints, and a second multi-source trajectory data set is obtained according to a matching result; filling space-time missing information in the second multi-source trajectory data set based on road network space-time constraints, and obtaining a third multi-source trajectory data set according to a filling result; the problem of fragmentation of the multi-source vehicle track is solved, and the completeness of vehicle track recovery is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a vehicle trajectory recovery method, a computer device, and a storage medium. Background Art

[0002] Vehicle trajectories are fine-grained representations of traffic flow. The recovery of complete vehicle trajectories is of great significance for understanding traffic travel demand, and provides important analysis and decision-making basis for downstream tasks such as traffic flow prediction, traffic travel planning and management.

[0003] Current vehicle trajectory recovery methods, such as VeTrac (Vehicle Tracking), STVTR (Spatio-Temporal Video Text Recognition), HMM (Hidden Markov Model), etc., are mostly based on single-source traffic perception data. However, the vehicle trajectory recovered based on single-source traffic perception data cannot cover the complete journey of a vehicle on the road network, resulting in blind spots in vehicle trajectory recovery. Summary of the invention

[0004] Based on this, it is necessary to provide a more complete vehicle trajectory recovery method, computer equipment and storage medium to address the above technical issues.

[0005] In a first aspect, the present application provides a vehicle trajectory recovery method. The method comprises: Construct single-source trajectory datasets for at least two types of traffic perception data, namely, GPS data, vehicle capture data, and toll station gate data; Performing fusion processing on each of the single-source trajectory data sets to obtain a first multi-source trajectory data set; Based on the spatiotemporal constraints of the road network, cross-data source matching is performed on the vehicle identities corresponding to the trajectory fragments in the first multi-source trajectory dataset, and a second multi-source trajectory dataset is obtained according to the matching results; The spatiotemporal missing information in the second multi-source trajectory dataset is filled based on the spatiotemporal constraints of the road network, and a third multi-source trajectory dataset is obtained according to the filling result.

[0006] In one embodiment, the single-source trajectory data sets are fused to obtain a first multi-source trajectory data set, including: Reading each of the single-source trajectory data sets to determine the source of the single-source trajectory data; In the case where it is determined that the source of the single-source trajectory data is GPS, the GPS data points with errors in the GPS data are matched to the corresponding road sections in the road network to obtain a format-converted GPS data set, and the format-converted GPS data set is merged with the other format-converted single-source trajectory data sets; or, When it is determined that the single-source trajectory data is from a road camera, identifying the identity and appearance features of the vehicle from the captured vehicle data; Calculate the similarity probability between every two vehicle images based on the identified vehicle identity and appearance features; Clustering the vehicle images with the similarity probability as a weight, forming trajectories from the clustered vehicle image groups, obtaining a vehicle capture data set after format conversion, and merging the vehicle capture data set after format conversion with other single-source trajectory data sets after format conversion; When it is determined that the source of the single-source trajectory data is a toll gate, the formats of the single-source trajectory data sets are unified before merging.

[0007] In one embodiment, when it is determined that the source of the single-source trajectory data is a road camera, the similarity probability between every two vehicle images is calculated according to the identified vehicle identity and appearance features, including: Determining the probability of identical identities based on the number of mismatched license plate numbers in the vehicle identities; Determining the appearance similarity probability according to the cosine product of the appearance features; Determine the probability of spatiotemporal coincidence based on the extreme degree of the derived vehicle speed; The similarity probability between every two vehicle images is determined according to the identity same probability, the appearance similarity probability, and the spatiotemporal coincidence probability.

[0008] In one of the embodiments, cross-data source matching is performed on vehicle identities corresponding to trajectory fragments in the first multi-source trajectory dataset based on road network spatiotemporal constraints, and a second multi-source trajectory dataset is obtained according to the matching results, including: Reading from the first multi-source trajectory data set a trajectory fragment group pair formed by combining every two first trajectory fragment groups with the same vehicle identity; Predicting a probability score that two of the first trajectory fragment groups in the trajectory fragment group pair belong to the same trajectory; clustering the first trajectory fragment groups using the probability scores as weights; The clustered first trajectory fragment groups are merged into a new second trajectory fragment group, and the second multi-source trajectory dataset is generated based on the second trajectory fragment group.

[0009] In one of the embodiments, predicting the probability scores that two of the first trajectory fragment groups in the trajectory fragment group pair belong to the same trajectory includes: Acquire a third trajectory fragment group from each single-source trajectory data set in a unified data structure; Randomly sampling pairs of the third trajectory fragment groups, and using whether the real identities of two third trajectory fragment groups in the trajectory fragment group pair match or not as a binary label, and generating a first training data set according to the trajectory fragment group pair and the binary label; Constructing a first spatiotemporal graph converter model, and performing self-supervised training on the first spatiotemporal graph converter model using the first training data set; Based on the trained first spatiotemporal graph converter model, a probability score that two of the first trajectory fragment groups in the trajectory fragment group pair belong to the same trajectory is predicted.

[0010] In one embodiment, filling the spatiotemporal missing information in the second multi-source trajectory dataset based on the spatiotemporal constraint of the road network, and obtaining the third multi-source trajectory dataset according to the filling result, comprises: reading a trajectory fragment group from the second multi-source trajectory dataset; Predicting the first spatiotemporal missing data of each of the trajectory fragment groups to obtain a complete trajectory of each of the trajectory fragment groups; A plurality of the complete trajectories are merged to obtain a complete trajectory dataset, and the third multi-source trajectory dataset is generated based on the complete trajectory dataset.

[0011] In one embodiment, predicting the first spatiotemporal missing data of each of the trajectory fragment groups to obtain a complete trajectory of each of the trajectory fragment groups includes: Randomly select the first trajectory fragment regarded as a complete trajectory from the cross-data source trajectory fragment group dataset; Randomly cutting off some track fragments from the first track fragments to generate second track fragments, and using the first track fragments as labels, generating a second training data set according to the second track fragments and the labels; Constructing a second spatiotemporal graph converter model, and performing self-supervised training on the second spatiotemporal graph converter model using the second training data set; Based on the trained second spatiotemporal graph converter model, the first spatiotemporal missing data of each of the trajectory fragment groups are predicted to obtain a complete trajectory of each of the trajectory fragment groups.

[0012] In one embodiment, after predicting the first spatiotemporal missing data of each of the trajectory fragment groups based on the trained second spatiotemporal graph converter model to obtain the complete trajectory of each of the trajectory fragment groups, the method further includes: Determine whether the complete trajectory dataset needs to be desensitized in downstream tasks; If the judgment is yes, continue to train the second spatiotemporal graph converter model, and gradually increase the spatiotemporal missing ratio of the input trajectory fragments until they are completely missing; Based on the currently trained second spatiotemporal graph sequence converter model, randomly predict and generate the second spatiotemporal missing data to obtain a new trajectory; A plurality of the new trajectories are merged to obtain a new trajectory dataset, and the new trajectory dataset is used to replace the complete trajectory dataset.

[0013] In a second aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in the first aspect are implemented.

[0014] In a third aspect, the present application further provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0015] The above-mentioned vehicle trajectory recovery method, computer equipment and storage medium respectively construct single-source trajectory data sets for at least two types of traffic perception data, namely GPS data, vehicle capture data and toll station gate data, to perform single-source vehicle trajectory recovery, and then fuse multiple single-source trajectory data sets, and then perform cross-data source matching and spatiotemporal completion on the fused multi-source trajectory data sets based on the spatiotemporal constraints of the road network to achieve complete vehicle trajectory recovery, solve the problem of multi-source vehicle trajectory fragmentation, and improve the integrity of vehicle trajectory recovery. It can then assist the transportation field to achieve more stable traffic flow prediction and provide a decision-making basis for urban transportation, management and planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A diagram showing an application environment of a vehicle trajectory recovery method in an embodiment; Figure 2 A schematic diagram of a process flow of a vehicle trajectory recovery method in one embodiment; Figure 3 is a flowchart of a method for acquiring a first multi-source trajectory data set in one embodiment; Figure 4 is a flowchart of a method for acquiring a second multi-source trajectory dataset in one embodiment; Figure 5 is a flowchart of a third multi-source trajectory data set acquisition method in one embodiment; Figure 6 is a flowchart of a third multi-source trajectory data set acquisition method in another embodiment; Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0018] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with general skills in the technical field to which this application belongs. The words "one", "a", "the", "these" and the like in this application do not indicate a quantitative limitation, and they may be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone. Generally, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0019] Most of the vehicle trajectory methods in related technologies are based on single-source traffic perception data. The recovered vehicle trajectory cannot cover the complete journey of a vehicle on the road network, resulting in blind spots in vehicle trajectory recovery. With the development and improvement of the intelligent perception system of the road network, traffic perception equipment has gradually become more abundant, such as GPS, road cameras, toll gates, etc., and the scale of collected data has exploded. Multi-source, large-scale traffic perception data makes it possible to recover vehicle trajectories.

[0020] However, recovering complete vehicle trajectories from multi-source traffic perception data is still challenging. Because vehicle trajectory data comes from many sources but has different structures, there are problems such as spatiotemporal missing, spatiotemporal noise, and vehicle identity ambiguity in different dimensions, resulting in the extraction of vehicle trajectories that are mostly fragmented, that is, a vehicle's complete journey on the road network is broken down into multiple short trajectories, which can only partially cover the journey. The existence of the fragmentation problem greatly limits the application scenarios of vehicle trajectories in smart transportation. For example, for traffic flow prediction, only short-term vehicle trajectory transfers can be mined in fragmented trajectory data, and the prediction effect can be improved for extremely short-term road demand increases, and its generalization is not strong. Therefore, how to solve the problem of reconstructing spatiotemporal graph sequences under massive fragmented trajectories under multi-source data is the key to recovering complete vehicle trajectories.

[0021] Based on the analysis of the above situation, the present application provides a vehicle trajectory recovery method, which can be applied to Figure 1 In the application environment shown. Among them, the traffic sensing device 1 communicates with the server 2 through the network. The data storage system 3 can store the data that the server 2 needs to process. The data storage system 3 can be integrated on the server 2, or it can be placed on the cloud or other network servers. Among them, the traffic sensing device 1 can be but is not limited to a vehicle GPS, a road camera, a toll station gate, etc. The server 2 can be implemented with an independent server or a server cluster composed of multiple servers. The server 2 receives the traffic sensing data collected by the traffic sensing device 1, and executes the vehicle trajectory recovery method based on the traffic sensing data.

[0022] Figure 2 The flowchart of the vehicle trajectory recovery method in this embodiment is shown in FIG. Figure 1 The server 2 in the example is used as an example to illustrate the process, including the following steps: Step S1, constructing a single-source trajectory data set for at least two types of traffic perception data, namely, GPS data, vehicle capture data, and toll station gate data.

[0023] Specifically, the received GPS data is processed into structured data, whose fields include vehicle ID, time, longitude and latitude, that is, The received vehicle capture data is processed into unstructured data, including time, camera ID, and image, that is, The received toll gate data is processed into structured data, whose fields include time, road section ID, and vehicle ID, that is, The meanings of the formula symbols are explained as follows: i: data record number.

[0024] N: number of data records; where N GPS is the number of GPS positioning records, N camis the number of pictures captured by the camera, N toll The number of charging records.

[0025] veh i : The vehicle corresponding to the i-th record.

[0026] t i : The timestamp of the i-th record.

[0027] lat i : The latitude of the ith position.

[0028] lon i : The longitude of the ith position.

[0029] cam i : The camera that captured the i-th image.

[0030] img i : The i-th captured image.

[0031] v i : The entrance and exit toll station corresponding to the i-th charge.

[0032] Step S2: fusing the single-source trajectory data sets to obtain a first multi-source trajectory data set.

[0033] Specifically, from each single-source trajectory dataset Extract vehicle trajectories in a unified format and merge them into a trajectory dataset (i.e., the first multi-source trajectory dataset), where Represents a vehicle trajectory.

[0034] Step S3, performing cross-data source matching on vehicle identities corresponding to trajectory fragments in the first multi-source trajectory dataset based on the road network spatiotemporal constraints, and obtaining a second multi-source trajectory dataset according to the matching results.

[0035] Specifically, by building and training a spatiotemporal graph converter model, the vehicle identities corresponding to the trajectory fragments in D' are matched across data sources using the spatiotemporal constraints of the road network to obtain a cross-data source trajectory fragment group dataset (i.e., the second multi-source trajectory dataset).

[0036] Step S4, filling the spatiotemporal missing information in the second multi-source trajectory dataset based on the spatiotemporal constraints of the road network, and obtaining a third multi-source trajectory dataset according to the filling result.

[0037] Specifically, by constructing and training a spatiotemporal graph converter model, the spatiotemporal constraints of the road network are used to fill the spatiotemporal gaps in the trajectory fragment group of S to generate a complete trajectory dataset (i.e., the third multi-source trajectory dataset).

[0038] Optionally, in some embodiments, the above method may further include: Step S5, saving the complete trajectory data and visualizing the results.

[0039] In the above steps S1 to S4, single-source trajectory data sets for at least two types of traffic sensing data, namely GPS data, vehicle capture data, and toll gate data, are constructed to restore single-source vehicle trajectories, and then multiple single-source trajectory data sets are fused, and then cross-data source matching and spatiotemporal completion are performed on the fused multi-source trajectory data sets based on the spatiotemporal constraints of the road network to achieve complete vehicle trajectory restoration, solve the problem of multi-source vehicle trajectory fragmentation, and improve the integrity of vehicle trajectory restoration. This can assist the transportation field in achieving more stable traffic flow predictions and provide decision-making basis for urban transportation, management, and planning.

[0040] In one embodiment, Figure 3 A flowchart of a method for obtaining a first multi-source trajectory dataset is provided. In step S2, each single-source trajectory dataset is fused to obtain a first multi-source trajectory dataset. Figure 3 The steps shown are as follows: Step S21, reading a single-source trajectory dataset.

[0041] Step S22, determine the source of the single-source trajectory data; when it is determined that the source of the single-source trajectory data is GPS, proceed to step S23; when it is determined that the source of the single-source trajectory data is a road camera, proceed to step S24; when it is determined that the source of the single-source trajectory data is a toll station gate, proceed to step S26.

[0042] Step S23, execute the map matching algorithm: use the HMM algorithm to transform the GPS data D GPS GPS data points with errors ( lat i , lon i ) matches the corresponding road segment in the road network v i , get the GPS data set after format conversion ; Go to step S26.

[0043] Step S24, executing computer vision algorithm: using ResNet algorithm to capture data from vehicles img i Identify the blurred vehicle identity veh i Appearance characteristics app i .

[0044] Step S25, execute the spatiotemporal constraint algorithm: calculate the similarity probability between every two vehicle images based on the identified vehicle identity and appearance features, cluster the vehicle images with the similarity probability as the weight, form a trajectory from the clustered vehicle image group, and obtain the vehicle capture data set after format conversion. Among them, the identity probability is determined based on the number of license plate number mismatches in the vehicle identity; the appearance similarity probability is determined based on the cosine product of the appearance features; the spatiotemporal coincidence probability is determined based on the extreme degree of the derived vehicle speed; and the similarity probability between every two vehicle images is determined based on the identity probability, appearance similarity probability, and spatiotemporal coincidence probability.

[0045] For example, computing for each pair of images ( img i , img j )of similar probability ρ(i,j)=P id (i,j)•P app (i,j)•P st (i,j), where the probability of identity being the same is P id Determined by the number of dislocations in the license plate number, the probability of similar appearance P app Determined by the cosine product of the appearance features, the spatiotemporal coincidence probability P st The extreme degree of derived vehicle speed is determined by the degree to which the vehicle speed value deviates from the expected range, which can be calculated from the single-source trajectory data. The GCN model is then used to cluster vehicle images with the ρ value between images as the weight, and the clustered image groups form trajectories to obtain the format-converted data set. ; Go to step S26.

[0046] Step S26: unify the formats of the single-source trajectory data sets and then merge them. Specifically, unify the formats of the trajectory data sets from each source and merge them into one trajectory data set. .

[0047] In the above steps S21 to S26, the multi-dimensional traffic perception data is taken into consideration according to the factors of vehicle trajectory recovery. According to the characteristics of each data source, map matching, computer vision, and spatiotemporal constraints are used to perform single-source vehicle trajectory recovery, and vehicle trajectories in a unified format are extracted and merged to obtain heterogeneous fused data (i.e., the first multi-source trajectory data set), which is conducive to recovering a more complete vehicle trajectory.

[0048] In one embodiment, Figure 4A flowchart of a method for obtaining a second multi-source trajectory dataset is provided. In step S3, based on the spatiotemporal constraints of the road network, the vehicle identities corresponding to the trajectory fragments in the first multi-source trajectory dataset are matched across data sources, and a second multi-source trajectory dataset is obtained according to the matching results. Figure 4 The steps shown are as follows: Step S31, reading from the first multi-source trajectory data set a trajectory fragment group pair formed by merging every two first trajectory fragment groups with the same vehicle identity. Specifically, the trajectories with the same vehicle identity in the data set D' are merged to form a trajectory fragment group , binary combinations form trajectory fragment pairs .in, l Represents the track fragment number.

[0049] Step S32, predicting the probability scores of the two first trajectory fragment groups in the trajectory fragment group pair belonging to the same trajectory. Specifically, the spatiotemporal graph sequence converter model M1 is used to predict the probability scores of each pair of trajectory fragment groups in the trajectory fragment group pair. The probability score Y(i, j) belonging to the same trajectory is used as the identity matching probability of the trajectory fragment group.

[0050] Step S33: cluster the first trajectory fragment groups using the probability score as the weight. Specifically, the graph convolution model is used to cluster the trajectory fragment groups using the probability score Y of the trajectory fragment groups belonging to the same trajectory as the weight to obtain M clusters. The cluster label is C i , so the clustering results are obtained by merging .

[0051] Step S34: merge the clustered first trajectory fragment groups into a new second trajectory fragment group, and generate a second multi-source trajectory dataset based on the second trajectory fragment group. Specifically, merge the clustered trajectory fragment groups into a new trajectory fragment group to obtain a cross-data source trajectory fragment group dataset. .

[0052] Wherein, before predicting the probability scores of the two first trajectory fragment groups in the trajectory fragment group pair belonging to the same trajectory in step S32, the following steps may also be included: Step S35, obtaining a third trajectory fragment group from each single-source trajectory dataset in a unified data structure. Specifically, construct each single-source trajectory dataset in a unified structure , from which track fragment groups are randomly merged or cut out .

[0053] Step S36, randomly sampling paired third track fragment groups, and using whether the real identities of the two third track fragment groups in the track fragment group pair match or not as a binary label, and generating a first training data set according to the track fragment group pair and the binary label. Specifically, randomly sampling paired track fragment groups and using whether their real identities match or not as a binary label, wherein the binary label is defined as follows:

[0054] In this way, a data set is formed , the first training data set is obtained by merging across data sources.

[0055] Step S37, construct a first spatiotemporal graph converter model, and use the first training data set to perform self-supervised training on the first spatiotemporal graph converter model. Specifically, construct a first spatiotemporal graph converter model M1, and use the first training data set of step S37 for self-supervised training until convergence, and save the optimal model parameters.

[0056] In one embodiment, Figure 5 A flowchart of a method for obtaining a third multi-source trajectory dataset is provided. In step S4, the spatiotemporal missing information in the second multi-source trajectory dataset is filled based on the spatiotemporal constraints of the road network, and a third multi-source trajectory dataset is obtained according to the filling result. Figure 5 The steps shown are as follows: Step S41, reading a trajectory fragment group from a second multi-source trajectory dataset. Specifically, reading a trajectory fragment group from a cross-data source trajectory fragment group dataset S .

[0057] Step S42: predict the first spatiotemporal missing data of each trajectory fragment group to obtain the complete trajectory of each trajectory fragment group. Specifically, the second spatiotemporal graph sequence converter model M2 is used to generate each trajectory fragment group. The missing space-time part in the complete trajectory T j .

[0058] Step S43, merging multiple complete trajectories to obtain a complete trajectory dataset, and generating a third multi-source trajectory dataset based on the complete trajectory dataset. Specifically, merging the complete trajectories generated by the second spatiotemporal graph sequence converter model M2 to obtain a complete trajectory dataset .

[0059] Wherein, in step S42, before predicting the first spatiotemporal missing data of each trajectory fragment group to obtain the complete trajectory of each trajectory fragment group, the following steps may also be included: Step S44: randomly select a first trajectory fragment regarded as a complete trajectory from the cross-data source trajectory fragment group dataset. Specifically, randomly select a trajectory fragment T from the cross-data source trajectory fragment group dataset S.i , and treat it as a complete trajectory.

[0060] Step S45, randomly cutting off some track fragments from the first track fragments to generate second track fragments, and using the first track fragments as labels, generating a second training data set according to the second track fragments and the labels. i Some of the track fragments, and the complete T i As labels, a second training dataset with time and space missing is formed.

[0061] Step S46, construct a second spatiotemporal graph converter model, and use the second training data set to perform self-supervised training on the second spatiotemporal graph converter model. Specifically, construct a second spatiotemporal graph sequence converter model M2, and use the second training data set of step S45 for self-supervised training until convergence, and save the optimal model parameters.

[0062] Above Figure 4 , Figure 5 The embodiment shown is based on the single-source trajectory extraction method of map matching, computer vision, and spatiotemporal constraints, combined with machine learning models such as converters and graph convolutional networks, and provides a method dedicated to vehicle trajectory recovery by fusing multi-source traffic perception data. Specifically, the heterogeneous data fusion part extracts vehicle trajectories in a unified format from each single-source trajectory data set through map matching algorithms, computer vision algorithms, and spatiotemporal constraint algorithms; the spatiotemporal graph sequence clustering part constructs and trains a spatiotemporal graph converter model, combines a graph convolution model, and uses the spatiotemporal constraints of the road network to match the vehicle identities corresponding to the trajectory fragments in D' across data sources; the spatiotemporal graph sequence generation part constructs and trains a spatiotemporal graph converter model, uses the spatiotemporal constraints of the road network to fill in the spatiotemporal gaps in the trajectory fragment group of S, and generates a complete trajectory. It effectively solves the problem of spatiotemporal graph sequence reconstruction under massive fragmented trajectories under multi-source data, and realizes the recovery of complete vehicle trajectories.

[0063] In one embodiment, Figure 6 A flowchart of another method for obtaining a third multi-source trajectory dataset is provided. Figure 5 On the basis of , in step S43, based on the trained second spatiotemporal graph sequence converter model M2, the first spatiotemporal missing data of each trajectory fragment group is predicted to obtain the complete trajectory of each trajectory fragment group, and the method further comprises the following steps: Step S47, determine whether the complete trajectory dataset D'' needs to be desensitized in the downstream task; if the determination is yes, proceed to step S48; otherwise, proceed to step S51.

[0064] Step S48, continue to train the second spatiotemporal graph converter model M2, and gradually increase the spatiotemporal missing ratio of the input trajectory fragments until they are completely missing. Specifically, train until convergence, and save the optimal model parameters.

[0065] Step S49: Based on the currently trained second spatiotemporal graph sequence converter model M2, randomly predict and generate second spatiotemporal missing data to obtain a new trajectory.

[0066] Step S50 , merging multiple new trajectories to obtain a new trajectory dataset, and replacing the complete trajectory dataset D′′ with the new trajectory dataset.

[0067] Step S51, saving the complete trajectory data and visualizing the results.

[0068] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store a single-source trajectory data set and / or a multi-source trajectory data set. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a vehicle trajectory recovery method is implemented.

[0069] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0070] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented: Construct single-source trajectory datasets for at least two types of traffic perception data, namely, GPS data, vehicle capture data, and toll station gate data; Performing fusion processing on each single-source trajectory data set to obtain a first multi-source trajectory data set; Based on the spatiotemporal constraints of the road network, cross-data source matching is performed on the vehicle identities corresponding to the trajectory fragments in the first multi-source trajectory dataset, and a second multi-source trajectory dataset is obtained according to the matching results; The spatiotemporal missing information in the second multi-source trajectory dataset is filled based on the spatiotemporal constraints of the road network, and a third multi-source trajectory dataset is obtained according to the filling result.

[0071] In addition, in combination with the vehicle trajectory recovery method provided in the above embodiments, a storage medium may be provided in this embodiment to implement the vehicle trajectory recovery method. The storage medium stores a computer program; when the computer program is executed by a processor, any vehicle trajectory recovery method in the above embodiments is implemented.

[0072] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: Construct single-source trajectory datasets for at least two types of traffic perception data, namely, GPS data, vehicle capture data, and toll station gate data; Performing fusion processing on each single-source trajectory data set to obtain a first multi-source trajectory data set; Based on the spatiotemporal constraints of the road network, cross-data source matching is performed on the vehicle identities corresponding to the trajectory fragments in the first multi-source trajectory dataset, and a second multi-source trajectory dataset is obtained according to the matching results; The spatiotemporal missing information in the second multi-source trajectory dataset is filled based on the spatiotemporal constraints of the road network, and a third multi-source trajectory dataset is obtained according to the filling result.

[0073] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0074] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0075] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A vehicle trajectory recovery method, characterized in that: include: Construct single-source trajectory datasets for at least two types of traffic perception data, namely, GPS data, vehicle capture data, and toll station gate data; Performing fusion processing on each of the single-source trajectory data sets to obtain a first multi-source trajectory data set; Based on the spatiotemporal constraints of the road network, cross-data source matching is performed on the vehicle identities corresponding to the trajectory fragments in the first multi-source trajectory dataset, and a second multi-source trajectory dataset is obtained according to the matching results; The spatiotemporal missing information in the second multi-source trajectory dataset is filled based on the spatiotemporal constraints of the road network, and a third multi-source trajectory dataset is obtained according to the filling result.

2. The vehicle trajectory recovery method according to claim 1, characterized in that: The single-source trajectory data sets are fused to obtain a first multi-source trajectory data set, including: Reading each of the single-source trajectory data sets to determine the source of the single-source trajectory data; In the case where it is determined that the source of the single-source trajectory data is GPS, the GPS data points with errors in the GPS data are matched to the corresponding road sections in the road network to obtain a format-converted GPS data set, and the format-converted GPS data set is merged with the other format-converted single-source trajectory data sets; or, When it is determined that the single-source trajectory data is from a road camera, identifying the identity and appearance features of the vehicle from the captured vehicle data; Calculate the similarity probability between every two vehicle images based on the identified vehicle identity and appearance features; Clustering the vehicle images with the similarity probability as a weight, forming trajectories from the clustered vehicle image groups, obtaining a vehicle capture data set after format conversion, and merging the vehicle capture data set after format conversion with other single-source trajectory data sets after format conversion; When it is determined that the source of the single-source trajectory data is a toll gate, the formats of the single-source trajectory data sets are unified before merging.

3. The vehicle trajectory recovery method according to claim 2, characterized in that: When it is determined that the source of the single-source trajectory data is a road camera, the similarity probability between each two vehicle images is calculated according to the identified vehicle identity and appearance features, including: Determining the probability of identical identities based on the number of mismatched license plate numbers in the vehicle identities; Determining the appearance similarity probability according to the cosine product of the appearance features; Determine the probability of spatiotemporal coincidence based on the extreme degree of the derived vehicle speed; The similarity probability between every two vehicle images is determined according to the identity same probability, the appearance similarity probability, and the spatiotemporal coincidence probability.

4. The vehicle trajectory recovery method according to claim 1, characterized in that: Based on the road network spatiotemporal constraints, cross-data source matching is performed on the vehicle identities corresponding to the trajectory fragments in the first multi-source trajectory dataset, and a second multi-source trajectory dataset is obtained according to the matching results, including: Reading from the first multi-source trajectory data set a trajectory fragment group pair formed by combining every two first trajectory fragment groups with the same vehicle identity; Predicting a probability score that two of the first trajectory fragment groups in the trajectory fragment group pair belong to the same trajectory; clustering the first trajectory fragment groups using the probability scores as weights; The clustered first trajectory fragment groups are merged into a new second trajectory fragment group, and the second multi-source trajectory dataset is generated based on the second trajectory fragment group.

5. The vehicle trajectory recovery method according to claim 4, characterized in that: Predicting the probability scores that two of the first trajectory fragment groups in the trajectory fragment group pair belong to the same trajectory includes: Acquire a third trajectory fragment group from each single-source trajectory data set in a unified data structure; Randomly sampling pairs of the third trajectory fragment groups, and using whether the real identities of two third trajectory fragment groups in the trajectory fragment group pair match or not as a binary label, and generating a first training data set according to the trajectory fragment group pair and the binary label; Constructing a first spatiotemporal graph converter model, and performing self-supervised training on the first spatiotemporal graph converter model using the first training data set; Based on the trained first spatiotemporal graph converter model, a probability score that two of the first trajectory fragment groups in the trajectory fragment group pair belong to the same trajectory is predicted.

6. The vehicle trajectory recovery method according to claim 1, characterized in that: Filling the spatiotemporal missing information in the second multi-source trajectory dataset based on the spatiotemporal constraints of the road network, and obtaining a third multi-source trajectory dataset according to the filling result, including: reading a trajectory fragment group from the second multi-source trajectory dataset; Predicting the first spatiotemporal missing data of each of the trajectory fragment groups to obtain a complete trajectory of each of the trajectory fragment groups; A plurality of the complete trajectories are merged to obtain a complete trajectory dataset, and the third multi-source trajectory dataset is generated based on the complete trajectory dataset.

7. The vehicle trajectory recovery method according to claim 6, characterized in that: Predicting the first spatiotemporal missing data of each of the trajectory fragment groups to obtain a complete trajectory of each of the trajectory fragment groups includes: Randomly select the first trajectory fragment regarded as a complete trajectory from the cross-data source trajectory fragment group dataset; Randomly cutting off some track fragments from the first track fragments to generate second track fragments, and using the first track fragments as labels, generating a second training data set according to the second track fragments and the labels; Constructing a second spatiotemporal graph converter model, and performing self-supervised training on the second spatiotemporal graph converter model using the second training data set; Based on the trained second spatiotemporal graph converter model, the first spatiotemporal missing data of each of the trajectory fragment groups are predicted to obtain a complete trajectory of each of the trajectory fragment groups.

8. The vehicle trajectory recovery method according to claim 7, characterized in that: After predicting the first spatiotemporal missing data of each of the trajectory fragment groups based on the trained second spatiotemporal graph converter model to obtain a complete trajectory of each of the trajectory fragment groups, the method further includes: Determine whether the complete trajectory dataset needs to be desensitized in downstream tasks; If the judgment is yes, continue to train the second spatiotemporal graph converter model, and gradually increase the spatiotemporal missing ratio of the input trajectory fragments until they are completely missing; Based on the currently trained second spatiotemporal graph sequence converter model, randomly predict and generate the second spatiotemporal missing data to obtain a new trajectory; A plurality of the new trajectories are merged to obtain a new trajectory dataset, and the new trajectory dataset is used to replace the complete trajectory dataset.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images